Perplexity’s new computer is another bet where users will need a lot of AI models

Starting this week, Perplexity subscribers will have new agent tools at their disposal.

Perplexity Computer, in the company’s words, “integrates all of today’s AI capabilities into one system.” Specifically, Perplexity says it is a computer user agent that can independently run complex workflows using 19 different AI models, and can also create subagents to handle specific problems.

The tool is currently only available in the company’s highest subscription tier, Perplexity Max, which costs $200 per month. Because it runs entirely in the cloud, it potentially avoids the security concerns of other agent tools such as OpenClaw.

TechCrunch hasn’t had a hands-on demo of the new tool, but example workflows on Perplexity’s website show processing tasks that involve gathering statistical, financial, and legal data. Creating an analysis. Then share your results as a finished website or visualization.

Perplexity invited the press to a background session with executives last week to discuss the product and lay out this year’s agenda. The event was scheduled to include a demonstration of the tool, but the company canceled the demo after a defect was discovered in the product hours before the event.

This tool represents an evolution of Perplexity. Perplexity made waves at the beginning of the AI ​​boom by wrapping frontier models into a familiar user interface, specifically a search engine-like response service. Then, last summer, we launched the Comet web browser. Competitors such as Google have changed their products to resemble those developed by Perplexity, which is both a compliment and a threat, one executive said.

The company is changing as the ecosystem changes. The company, one of the first AI companies to serve ads, announced last week that it was eroding user confidence in the accuracy of answers and announced it was abandoning the business late last year. But Perplexity’s total user base (tens of millions of users) pales in comparison to that of OpenAI, which claims to have 800 million weekly users and began testing ads on ChatGPT this year.

Now, Perplexity executives say they are aiming for a more boutique audience, offering products that help people who make “GDP-shaping decisions.” Executives who attended the briefing spoke on condition of anonymity, explaining that the company prioritizes enterprise subscriptions, especially when it comes to in-depth research.

“You never hear us talking about MAU because we’re not really on a mission to acquire as many users as possible,” one executive said.

Perplexity recently released a new benchmark for complex research tasks called Draco. There (not surprisingly) its unique in-depth research service beats out competitors like Gemini.

Perplexity says it no longer relies on third-party APIs for web indexing and has introduced its own AI-optimized search API. However, the company is committed to packaging its frontier model for a consumer-friendly user experience, arguing that there is value in coordinating multiple third-party LLMs to get the most cost-effective and accurate answers to queries.

“Multi-models are the future,” claimed one Perplexity executive. In their view, the model is not about commoditizing, but about specializing. The company found that users frequently switched between models to get the results they were looking for. December 2025 queries for visual output were most frequently sent to Gemini Flash, software engineering was done with Claude Sonnet 4.5, and medical research was done with GPT-5.1.

If one LLM excels at coding tasks and another excels at drafting marketing copy, Perplexity’s software automatically selects the ideal LLM. Another example, executives said, is Perplexity’s own modification of a Chinese-made open source LLM to answer queries more cheaply, a technique the company was accused of hiding from customers last year. However, this technique, which is performed transparently, could prove to be an efficient way to optimize LLM queries.

The company also offers users the opportunity to query multiple models simultaneously with a feature called Model Council. However, the unit economics of providing multiple queries for a flat subscription fee are not entirely clear.

Still, executives argue that Perplexity can remain competitive by allocating tokens to the model that best fits their purpose, as it has no expensive infrastructure projects on its books and high margins on user fees.

The Perplexity Comet browser is coming to iOS next month, and the company is planning an “Ask” developer conference in San Francisco on March 11 to encourage third-party usage of the API.

One executive said that every morning, instead of looking at the previous day’s query count, he now looks at the latest revenue metrics. At least some customers have noticed a new focus on revenue on Perplexity’s subreddit, featuring frequent complaints about new rate limits at free and subscription product levels.

However, executives who attended the briefing dismissed these complaints and said, “The discussion about reforming the free tier and limiting fees is completely false.”